The scariest problems about big existential threats to humanity is people not taking them seriously. Not understanding that we aren't invincible.
3 years ago | 9
Not enough people have thought about the existentially paralyzing idea that we could just fail.
3 years ago | 2
i’ve been thinking this for a while now. it’s cool to see my loose thoughts as an actually organized video. well done
3 years ago | 0
I think it also stems from how we look at history. We were the one who figured it out. But time and time again you see millions of people wiped out as a footnote in history (in a “meaningless” event) But they are not here to tell their story.
3 years ago | 1
Weird how i never saw these posts until i went deliberately looking on your channel for links to resources, but i get shown posts from channels im not even subbed to all the time. This is clearly the biggest danger AI poses to us! On that note: does anyone have any recommendations for good sources to learn about AI more thoroughly? (I'm a maths grad if that changes what's best)
3 years ago | 3
`function AllOrNothing(objectivefunctions) { var j = Infinity; for (var i = 0; i < objectivefunctions.length; i++) { if (objectivefunctions[i] < j) { j = Math.min(j,objectivefunctions[i]); } } return j; }` I have no idea if this will work, but this allows you to stack multiple proxies which makes it harder for the AI to hack its reward by making the attack surface smaller. I have been sitting on this for a while since I was afraid of being ridiculed for a bad idea. If you find this promising please share this far and wide.
1 year ago | 0
What if AI could only plan in steps, every once in awhile asking a human if what it's doing Is correct and happily resetting it's objective in whatever way the human wants (in a certain way having it's goal be to maximize a function than have it changed, or not, by a human with only the human deciding whether it changes it or not) That way humanity can never be killed by the ai tho I can see a future where the ai kinda enslaves humans to just press the (keep going button every week or so)
2 years ago | 0
Robert Miles AI Safety
I get frustrated sometimes. Here's a rant about it, on my second channel.
3 years ago | [YT] | 93